CRE CRE Sensors, Vision & Perception 2 — Questions and Answers
Question 1: Camera calibration in robot vision primarily determines:
- The camera's color balance settings
- Intrinsic parameters such as focal length and lens distortion coefficients (Correct answer)
- The optimal shutter speed for the scene
- The image file compression ratio
Correct answer: Intrinsic parameters such as focal length and lens distortion coefficients
Calibration estimates intrinsic parameters (focal length, principal point, distortion coefficients) that map pixel coordinates to accurate 3D measurements.
Question 2: Stereo vision systems estimate scene depth by:
- Measuring laser pulse time of flight
- Computing the disparity between matching points in two horizontally offset cameras (Correct answer)
- Analyzing color gradient magnitudes in a single image
- Projecting structured infrared dot patterns
Correct answer: Computing the disparity between matching points in two horizontally offset cameras
Stereo vision triangulates 3D depth from the horizontal pixel offset (disparity) of corresponding features viewed by two cameras with a known baseline.
Question 3: In robot vision, a homography matrix describes:
- The color mapping between two different cameras
- A projective transformation relating corresponding points on two planes (Correct answer)
- The lens focal length in pixel units
- The stereo camera baseline length
Correct answer: A projective transformation relating corresponding points on two planes
A homography is a 3×3 projective transformation that maps every point on one planar surface to its corresponding location in another camera view.
Question 4: The RANSAC algorithm is applied in computer vision to:
- Compress image data for network transmission
- Robustly estimate model parameters in the presence of data outliers (Correct answer)
- Calibrate camera intrinsic parameters
- Detect color blobs in cluttered scenes
Correct answer: Robustly estimate model parameters in the presence of data outliers
RANSAC iteratively fits models to minimal random subsets of data, identifying inliers and discarding outlier measurements to find a robust solution.
Question 5: Point cloud data from 3D depth sensors is typically processed using:
- Fast Fourier Transform libraries
- The Point Cloud Library (PCL) (Correct answer)
- Standard 2D image convolution filters
- Relational database SQL queries
Correct answer: The Point Cloud Library (PCL)
PCL provides specialized algorithms for filtering, segmentation, surface reconstruction, and feature extraction tailored to 3D point cloud data.
Question 6: Convolutional Neural Networks (CNNs) are primarily applied in robot vision for:
- Real-time joint torque control
- Image classification and object detection tasks (Correct answer)
- Kinematic path planning computations
- Encoder signal noise filtering
Correct answer: Image classification and object detection tasks
CNNs learn hierarchical feature representations from images and are widely used to classify scenes and detect or localize objects in robotic perception pipelines.
CRE CRE Sensors, Vision & Perception 2
Camera calibration in robot vision primarily determines: